CN108897619B - Multi-level resource flexible configuration method for super computer - Google Patents
Multi-level resource flexible configuration method for super computer Download PDFInfo
- Publication number
- CN108897619B CN108897619B CN201810680674.9A CN201810680674A CN108897619B CN 108897619 B CN108897619 B CN 108897619B CN 201810680674 A CN201810680674 A CN 201810680674A CN 108897619 B CN108897619 B CN 108897619B
- Authority
- CN
- China
- Prior art keywords
- time
- breakpoints
- job
- expected time
- supercomputer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/5038—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the execution order of a plurality of tasks, e.g. taking priority or time dependency constraints into consideration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5011—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals
- G06F9/5016—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals the resource being the memory
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Debugging And Monitoring (AREA)
- Multi Processors (AREA)
Abstract
The invention relates to a multi-level resource flexible configuration method for a super computer, which comprises the following steps:assigning P of supercomputers to jobs0A node; the calculation is performing N tasks T1,T2,...,TNReach M breakpoints in the process of { B } {1,B2,...,BMCorresponding initial expected time ofComputing when executing task TjLate arrival task TjAnd task Tj+1Break point B in betweeniThe actual time of (c)And the initial expected timeDifference of (2)When in useThen, for the remaining N-j unexecuted tasks { T }j+1,Tj+2,...,TNAllocate P1One compute node and recalculate to the remaining M-i breakpoints Bi+1,Bi+2,...,BMCorresponding first corrected expected time of
Description
Technical Field
The invention relates to a multi-level resource flexible configuration method for a super computer.
Background
The supercomputer is a computer which is formed by combining a plurality of computing nodes and can perform large-scale computation or data processing in parallel, is also called as a parallel computer, is the computer with the strongest function, the fastest operation and the largest storage capacity, is mainly used for the national high-tech field and the advanced technical research, and is an important embodiment of the national science and technology development level and the comprehensive national force.
At present, when a user submits a job to a supercomputer, various required resources, such as a storage space, a node number, a core number and the like of the supercomputer required for running the job, need to be specified by the user. In general, the user estimates the required resources based on experience or the results of a small number of data commissioning, and thus the bias is often large. If the requested resources are insufficient, the operation may be terminated due to timeout, overflow and the like, and a desired result cannot be obtained; however, if the resources requested are excessive, the user is charged with additional cost and valuable recalculation computing power is wasted. Therefore, how to specify a proper amount of resources for a job when the job is submitted and run becomes an urgent problem to be solved.
Disclosure of Invention
In order to solve the technical problem, the invention provides a multi-level resource flexible configuration method for a supercomputer, which comprises the following steps:
step S100, obtaining a job, wherein the job comprises N tasks { T }1,T2,...,TNAnd M breakpoints B respectively arranged between the tasks1,B2,...,BM}; assigning P of supercomputer to the job0A node; the calculation is performing N tasks T1,T2,...,TNReach M breakpoints in the process of { B } {1,B2,...,BMCorresponding initial expected time ofWherein N, M and P0Are all natural numbers, and M is more than N;
step S200, calculating the task T when executingjLate arrival task TjAnd task Tj+1Break point B in betweeniThe actual time of (c)And the initial expected timeDifference of (2)
Detailed Description
The present invention will be described in further detail in order to make the objects, technical solutions and advantages of the present invention more apparent. This description is made by way of example and not limitation to specific embodiments consistent with the principles of the invention, the description being in sufficient detail to enable those skilled in the art to practice the invention, other embodiments may be utilized and the structure of various elements may be changed and/or substituted without departing from the scope and spirit of the invention. The following detailed description is, therefore, not to be taken in a limiting sense.
One embodiment of the invention provides a multi-level resource flexible configuration method for a supercomputer, wherein the supercomputer is selected from a Tianhe supercomputer, in particular a Tianhe series supercomputer such as TH-1, TH-1A, TH-2, and the series supercomputer generally receives and executes a job in the form of a script file, wherein the script file at least provides parameters such as a job submission mode, a calculation partition, a node number, a core number, a task script file absolute file path and the like, and the submission form of reference is "yhbatch-N N1-p P1-N1xxx.bat", wherein N1 is the node number, and the data type is integer; p1 is the name of the partition, and the data type is a character string; n1 is the number of cores, and the data type is integer; bat is task script file name, data type is string, specifically, the configuration method includes the following steps:
step S100, obtaining a job through a script file, wherein the job comprises N tasks { T }1,T2,...,TNAnd M breakpoints B respectively arranged between the tasks1,B2,...,BMThe task can be any suitable software or a program for executing specific processing, the task has an interface for inputting data, the result of the input data after processing is used as output data, the result data output by the previous task is used as the input data of the next task, and the result of the task is obtained after the last task is executed, namely the execution of the task is finished; the breakpoint is arranged after one or more of the first N-1 tasks, the operation is temporarily suspended at the breakpoint, and the next task is continued after the execution progress of the operation is evaluated; assigning P of supercomputer to the job after obtaining script file0Each node generally comprises a plurality of computing cores, for example, 4-28 cores, and in the Tianhe super computer, computing resources are generally distributed by taking the node as a unit; then, before the job is executed, the N tasks are calculated to be executed { T }1,T2,...,TNReach M breakpoints in the process of { B } {1,B2,...,BMCorresponding initial expected time ofAnd the initial expected job run time required to complete the job, i.e., all N tasksWherein N, M and P0All are natural numbers, and M > N, the calculated initial expected timeAnd an initial expected job run timeAfter storing, the method is used for evaluating the execution progress of the job in the following steps;
step S200, after the job is submitted to the super computer to be executed, in the task TjAnd task Tj+1Breakpoint BiInterrupt processing, obtaining the operation from the start of execution to the breakpoint BiActual running time of the processCalculating the actual timeCorresponding to the initial expected timeDifference between them
Step S300, whenThen, for the remaining N-j unexecuted tasks { T }j+1,Tj+2,...,TNAllocate P1One compute node and recalculate to the remaining M-i breakpoints Bi+1,Bi+2,...,BMCorresponding first corrected expected time ofWhere | Δ t1Is Deltat1Absolute value of, TH1To set the threshold value, TH1May be any suitable value, typically not more than 10, preferably not more than 5, for example not more than 4, 3, 2, 1, 0.5, 0.3, 0.2, 0.1 etc.
In a preferred embodiment, in step S300, when Δ t is greater than or equal to1When > 0, P1=(1+w)×P0First corrected expected timeIn (1)Wherein M is more than or equal to i +1 and less than or equal to M,thus as i increases, w gets closer to A1Thereby providing more resources to complete the job as soon as possible within the expected time.
In a preferred embodiment, in step S300, when Δ t is greater than or equal to1When the ratio is less than or equal to 0, P1=(1-w)×P0First corrected expected timeIn (1)Wherein M is more than or equal to i +1 and less than or equal to M,this allows for the release of excess resources as early as possible to obtain completion in the right time and to save costs as much as possible.
Similarly, when task T is executedj+yLate arrival task Tj+yAnd task Tj+y+1Break point B in betweeni+xWhen in treatment, y is more than or equal to 1 and less than N-j, x is more than or equal to 1 and less than M-i, and the actual time is calculatedAnd the initial expected timeDifference of (2)When in useThen, for the remaining N-j-y tasks { T }j+y+1,Tj+y+2,...,TNAllocate PxOne compute node and recalculate to the remaining M-i-x breakpoints { B }i+x+1,Bi+x+2,...,BMCorresponding corrected expected time of xthWherein when | Δ txIs DeltatxAbsolute value of, THxTo set the threshold value, THxCan be connected with TH1The same or different. TH allows for more resources to be needed to correct the bias the later the program is, and thereforexPreferably less than TH1So as to sensitively and timely start the resource allocation correction process and complete the operation on time.
In some cases, in step S200, the number P of nodes that need to be allocated for the current operation of the job is calculated according to the algorithm complexity of the job and the data amount of the current operation, or according to the historical operation result of the job and the data amount of the current operation0And calculating the number of tasks in executing N { T }1,T2,...,TNReach M breakpoints in the process of { B } {1,B2,...,BMCorresponding initial expected time ofFor example, the historical run results include the job at different data volumes { D1,D2,...,DLAnd number of different nodes { P }1,P2,...,PKRun time under the conditions of }And run to M breakpoints { B1,B2,...,BMCorresponding time of }WhereinAndrespectively indicate the data quantity D of the operationaAnd the number of nodes PbRun time and run arrival breakpoint B under the condition of (c)iThe time of (d); accordingly, the data amount D of the current operation is calculated according to the operationcAnd a desired run time tcSearching and selecting from the historical operation resultsCorresponding number of nodes PbThe number P of the nodes needing to be distributed as the operation of the operation at this time0SelectingReach M breakpoints as runs B1,B2,...,BMCorresponding initial expected time ofWherein D isaAnd DcClosest to and equal to or greater than DcAt the same timeAnd tcClosest to and not more than tc。
In other cases, when the number P of nodes to be allocated for the current operation of the job cannot be found in the historical operation result according to the data size and the expected operation time of the current operation of the job0And run to M breakpoints { B1,B2,...,BMCorresponding toInitial expected timeIn this case, various known interpolation methods can be used to determine the number P of nodes to be allocated in the current operation of the job0And run to M breakpoints { B1,B2,...,BMCorresponding initial expected time of
In some cases, at allocation P0At the same time, according to N tasks { T }1,T2,...,TNAllocating memory and storage space according to the requirement of the unit; and in distribution P1While computing nodes, according to the remaining N-j unexecuted tasks { Tj+1,Tj+2,...,TNAllocating memory and storage space according to the requirement of the unit; and in distribution PxWhile computing the node, according to the rest N-j-y tasks { T }j+y+1,Tj+y+2,...,TNAllocating memory and storage space according to the requirement of the unit; wherein any known method may be employed to allocate memory and storage space. Preferably, the storage space is increased when the allocated storage space occupancy exceeds 85%, preferably exceeds 75%, more preferably exceeds 70%.
By using the method of the invention, before the operation is executed, the proper resources can be initially allocated to the operation as accurately as possible according to the historical data; during the execution of the operation, the resources can be dynamically allocated or recovered according to the actual execution condition of the operation, namely the condition of exceeding or falling below the expected progress, so that the flexible or elastic configuration of the resources of the super computer at multiple levels is realized, and the occupation of the resources of the super computer is reduced as much as possible under the condition of fully ensuring the timely completion of the operation.
Moreover, other implementations of the invention will be apparent to those skilled in the art from consideration of the specification of the invention disclosed herein. The embodiments and/or aspects of the embodiments can be used in the systems and methods of the present invention alone or in any combination. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.
Claims (5)
1. A method for flexible configuration of multi-tier resources for a supercomputer, comprising the steps of:
step S100, obtaining a job, wherein the job comprises N tasks { T }1,T2,...,TNAnd M breakpoints B respectively arranged between the tasks1,B2,...,BM}; assigning P of supercomputer to the job0A node; the calculation is performing N tasks T1,T2,...,TNReach M breakpoints in the process of { B } {1,B2,...,BMCorresponding initial expected time ofWherein N, M and P0Are all natural numbers, and M is more than N;
step S200, calculating the task T when executingjLate arrival task TjAnd task Tj+1Break point B in betweeniThe actual time of (c)And the initial expected timeDifference of (2)Calculating the number P of nodes required to be allocated for the current operation of the operation according to the algorithm complexity of the operation and the data volume of the current operation or according to the historical operation result of the operation and the data volume of the current operation0And calculating the number of tasks in executing N { T }1,T2,...,TNReach M breakpoints in the process of { B } {1,B2,...,BMCorresponding initial expected time ofWherein the historical operating result comprises that the job has different data volumes { D1,D2,...,DLAnd number of different nodes { P }1,P2,...,PKRun time under the conditions of }And run to M breakpoints { B1,B2,...,BMCorresponding time of }WhereinAndrespectively indicate the data quantity D of the operationaAnd the number of nodes PbRun time and run arrival breakpoint B under the condition of (c)iThe time of (d); accordingly, the data amount D of the current operation is calculated according to the operationcAnd a desired run time tcSearching and selecting from the historical operation resultsCorresponding number of nodes PbThe number P of the nodes needing to be distributed as the operation of the operation at this time0SelectingReach M breakpoints as runs B1,B2,...,BMCorresponding initial expected time ofWherein D isaAnd DcClosest to and equal to or greater than DcAt the same timeAnd tcClosest to and not more than tc;
2. The multi-tier resource flexible configuration method for a supercomputer according to claim 1, wherein, in step S300, TH1Is a set threshold value not exceeding 5.
3. The multi-level resource flexible configuration method for a supercomputer according to claim 1, wherein when the number P of nodes to be allocated for the current operation of the job cannot be found in the historical operation result according to the data amount and expected operation time of the current operation of the job0And run to M breakpoints { B1,B2,...,BMCorresponding initial expected time ofThen, the number P of the nodes needing to be distributed in the current operation of the operation is obtained by adopting an interpolation method0And run to M breakpoints { B1,B2,...,BMCorresponding initial expected time of
5. the multi-tier resource flexible configuration method for a supercomputer according to any one of claims 1 to 3, wherein in step S300, when Δ t is reached1When the ratio is less than or equal to 0, P1=(1-w)×P0First corrected expected timeIn (1)Wherein M is more than or equal to i +1 and less than or equal to M,
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010258601.8A CN111475297B (en) | 2018-06-27 | 2018-06-27 | Flexible operation configuration method |
CN201810680674.9A CN108897619B (en) | 2018-06-27 | 2018-06-27 | Multi-level resource flexible configuration method for super computer |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810680674.9A CN108897619B (en) | 2018-06-27 | 2018-06-27 | Multi-level resource flexible configuration method for super computer |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010258601.8A Division CN111475297B (en) | 2018-06-27 | 2018-06-27 | Flexible operation configuration method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108897619A CN108897619A (en) | 2018-11-27 |
CN108897619B true CN108897619B (en) | 2020-05-05 |
Family
ID=64346784
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810680674.9A Active CN108897619B (en) | 2018-06-27 | 2018-06-27 | Multi-level resource flexible configuration method for super computer |
CN202010258601.8A Active CN111475297B (en) | 2018-06-27 | 2018-06-27 | Flexible operation configuration method |
Family Applications After (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010258601.8A Active CN111475297B (en) | 2018-06-27 | 2018-06-27 | Flexible operation configuration method |
Country Status (1)
Country | Link |
---|---|
CN (2) | CN108897619B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110083449B (en) * | 2019-04-08 | 2020-04-28 | 清华大学 | Method and device for dynamically allocating memory and processor and computing module |
CN114020443B (en) * | 2022-01-05 | 2022-03-18 | 国家超级计算天津中心 | Supercomputer resource scheduling method, electronic device and medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102906696A (en) * | 2010-03-26 | 2013-01-30 | 维尔图尔梅特里克斯公司 | Fine grain performance resource management of computer systems |
CN106254058A (en) * | 2015-06-12 | 2016-12-21 | 华为技术有限公司 | A kind of method and device of the frequency adjusting server |
CN107045456A (en) * | 2016-02-05 | 2017-08-15 | 华为技术有限公司 | A kind of resource allocation methods and explorer |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8185422B2 (en) * | 2006-07-31 | 2012-05-22 | Accenture Global Services Limited | Work allocation model |
US9262216B2 (en) * | 2012-02-14 | 2016-02-16 | Microsoft Technologies Licensing, LLC | Computing cluster with latency control |
CN104252391B (en) * | 2013-06-28 | 2017-09-12 | 国际商业机器公司 | Method and apparatus for managing multiple operations in distributed computing system |
US9417928B2 (en) * | 2014-12-24 | 2016-08-16 | International Business Machines Corporation | Energy efficient supercomputer job allocation |
CN104536770A (en) * | 2015-01-28 | 2015-04-22 | 浪潮电子信息产业股份有限公司 | Job submitting and restoring method capable of supporting break restoration of concurrent jobs |
CN105808334B (en) * | 2016-03-04 | 2016-12-28 | 山东大学 | A kind of short optimization of job system and method for MapReduce based on resource reuse |
-
2018
- 2018-06-27 CN CN201810680674.9A patent/CN108897619B/en active Active
- 2018-06-27 CN CN202010258601.8A patent/CN111475297B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102906696A (en) * | 2010-03-26 | 2013-01-30 | 维尔图尔梅特里克斯公司 | Fine grain performance resource management of computer systems |
CN106254058A (en) * | 2015-06-12 | 2016-12-21 | 华为技术有限公司 | A kind of method and device of the frequency adjusting server |
CN107045456A (en) * | 2016-02-05 | 2017-08-15 | 华为技术有限公司 | A kind of resource allocation methods and explorer |
Also Published As
Publication number | Publication date |
---|---|
CN111475297A (en) | 2020-07-31 |
CN111475297B (en) | 2023-04-07 |
CN108897619A (en) | 2018-11-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10733026B2 (en) | Automated workflow selection | |
CN111381950B (en) | Multi-copy-based task scheduling method and system for edge computing environment | |
CN104765640B (en) | A kind of intelligent Service dispatching method | |
CN108205469B (en) | MapReduce-based resource allocation method and server | |
CN110347515B (en) | Resource optimization allocation method suitable for edge computing environment | |
CN108897619B (en) | Multi-level resource flexible configuration method for super computer | |
CN108427602B (en) | Distributed computing task cooperative scheduling method and device | |
CN104199739A (en) | Speculation type Hadoop scheduling method based on load balancing | |
CN115237580A (en) | Intelligent calculation-oriented flow parallel training self-adaptive adjustment system and method | |
CN111176810B (en) | Meteorological hydrology data processing scheduling system based on priority | |
CN115586961A (en) | AI platform computing resource task scheduling method, device and medium | |
CN113448714B (en) | Computing resource control system based on cloud platform | |
CN108108242B (en) | Storage layer intelligent distribution control method based on big data | |
CN108139929B (en) | Task scheduling apparatus and method for scheduling a plurality of tasks | |
CN111736959B (en) | Spark task scheduling method considering data affinity under heterogeneous cluster | |
CN113452546A (en) | Dynamic quality of service management for deep learning training communications | |
CN117331668A (en) | Job scheduling method, device, equipment and storage medium | |
CN116915869A (en) | Cloud edge cooperation-based time delay sensitive intelligent service quick response method | |
JP2006195985A (en) | Method for controlling resource utilization rate and computer system | |
Banicescu et al. | Towards the robustness of dynamic loop scheduling on large-scale heterogeneous distributed systems | |
CN112052087B (en) | Deep learning training system and method for dynamic resource adjustment and migration | |
CN111178529B (en) | Data processing method and device, electronic equipment and readable storage medium | |
CN112015539A (en) | Task allocation method, device and computer storage medium | |
KR101470695B1 (en) | Method and system of biogeography based optimization for grid computing scheduling | |
CN115280286A (en) | Dynamic allocation and reallocation of learning model computing resources |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |